Roadmap

rubberduck — MotherDuck-shaped hypertenant OLAP on Cloudflare.

P0 — now

Blog-ready reference. Live Parquet SQL on Workers. Measured benchmarks. Local proof that placement does not move data.

P1 — credible for one team

1. Per-org API keys (prod profile replaces shared demo token)
2. Scoped R2 credentials per org
3. Native graduation on the hosted path (same Parquet prefix)
4. Catalog + ingest conventions
5. Timeouts / result limits / Arrow
6. CI smoke + deploy runbook

Exit: third party deploys from docs; OOM-class queries reject cleanly or land on native.

P2 — something people pay for

SDKs / wire protocol, usage metering, admin UI, cells for noisy neighbors, SSO and a trust story. Optional CH-shaped ingest as a separate tier if customers need a firehose.

Scope boundaries

Enterprise Snowflake-scale warehouses, sub-20ms cold Workers, and Spark-grade ETL inside a duckling are out of scope — batch into R2 and query from rubberduck, or use a warehouse tier built for that job.